error when using multi-gpu option for trainingNetwork

Question

Wonsup Lee 2018-8-8

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/414024-error-when-using-multi-gpu-option-for-trainingnetwork

编辑： Wonsup Lee 2019-1-29

I've stuck with this problem for more than a month. In 'trainingOptions', 'gpu' is working for 'ExcutionEnvironment', but 'multi-gpu' is not working and showing the following error!

Parallel training of recurrent networks is not supported. 'ExecutionEnvironment' value in trainingOptions function must be 'auto', 'gpu', or 'cpu'.

I have two GPUs. If I check 'gpuDevice', I can see both GPUs. Also, I have Parallel Computing Toolbox, but I don't know what should I change in that toolbox in order to use the multi-gpu option.

Thank you so much in advance for your effort to help my problem!! W

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Joss Knight 2018-8-8

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/414024-error-when-using-multi-gpu-option-for-trainingnetwork#answer_332084

You have a recurrent network. So you have a network with a SequenceInputLayer and at least one LSTM layer. For these networks, multi-gpu training is not yet supported. Sorry.